Every Tool Just Added 'Agent Mode,' and Now I Know Why
Suddenly every app has an 'agent mode' button. It's not a coincidence; it's the same capability maturing in managed infrastructure and trickling into products all at once.
Suddenly every app has an 'agent mode' button. It's not a coincidence; it's the same capability maturing in managed infrastructure and trickling into products all at once.
The infrastructure arrived in September 2026. OpenAI's Agents API (September 10) exposed the harness behind long-running agents: context management, tool use, subagent coordination, and durable sessions where work persists across runs. AWS followed on September 18 with Bedrock AgentCore Runtime, a serverless microVM built for bursty, isolated agent execution.
That's why your writing app and your CRM both suddenly 'do agents.' They're standing on the same managed runtimes, which means the hard parts, persistence, isolation, cost control, are someone else's problem now.
The risk is expectation. An 'agent' that books your flight is impressive until it books the wrong one and there's no undo. The capability is real; the guardrails are still catching up, and that gap is where the early failures will happen.
Expect 'agent mode' to become as common as 'dark mode' was a few years ago. The differentiator won't be having it, it'll be whether it can be trusted with the consequences. Watch for the undo button before you let it act.